terramatch

TerraMatch

Find Earth’s next training ground.

TerraMatch is a NASA Space Apps Challenge 2026 project for exploring planetary terrain and comparing Earth analog candidates. The application contains an interactive Earth/Mars/Moon globe, an animated cargo-lander deployment, an actual controllable six-legged 3D explorer, and evidence-backed terrain comparisons. The application UI is implemented in HTML/CSS/ES modules, with Three.js vendored locally. Generated design images are not used as functioning UI.

Run locally

Requires Node.js 22 or later. No production dependency installation is needed.

npm start

Open http://127.0.0.1:4173 in a browser with WebGL 2 and hardware acceleration. The static application is in dist/. It can be served from any ordinary static web host. Planet textures, the 3D runtime, photographs, ten reference DEM subsets and the global land mask are served locally. There is no API key requirement. New manual/automatic searches require internet access to public terrain tiles; place-name searches additionally use Open-Meteo / GeoNames. Bundled reference comparisons work without those services. Use HTTPS or localhost for source hashes.

Use the application

1. Choose Mars or Moon in Explore, then select a target region. Mars includes Jezero, Gale / Bradbury Landing and Eberswalde delta area; Moon includes Shackleton, Cabeus crater center and Nobile crater center.
2. In Manual search, submit a place name or decimal latitude, longitude. Alternatively choose Pick on globe, rotate Earth, and click a location. Analyze this area fetches a new 32 × 32 km elevation window. Its source tile hashes, land screening, charts and provenance remain available.
3. In Automatic search, select a regional or global sample, or draw a region by clicking two opposite globe corners. Choose a budget of 12, 24 or 48 windows and start the survey. Actual new coordinate windows are fetched, analyzed and compared with the target. Progress, failure/exclusion counts and cancellation with partial results are supported. A global sample is not exhaustive.
4. Select slope and/or relief and run Rank loaded candidates. Results include bundled references and eligible on-demand windows. New candidates show their manual/automatic origin. The latest 24 valid search windows are cached locally.
5. Choose Explore in 3D. Watch the lander arrive, its door open, the ramp unfold, and the hexapod deploy. Skip arrival starts exploration immediately. The low TerraMatch Cargo Lander, powered by Wh0Knows, has four terrain-grounded legs, two side-opening cargo doors and a wide folding ramp. Its 15-second arrival includes engine exhaust, touchdown compression and fading surface dust; lunar grains use a separate ballistic visual style. The robot walks down the same ramp geometry used for foot contact, then transfers to keyboard control. This is an original fictional vehicle and illustrative arrival, rather than a validated spacecraft/soil simulation.
6. W/S move; A/D turn; Shift accelerates navigation. Drag to orbit the follow camera; scroll to zoom. E inspects terrain, C opens the Earth comparison sidebar, and Escape pauses. Touch movement controls appear on coarse-pointer devices.
7. Switch Surface / Slope / Elevation layers. Inspect and save geolocated terrain observations with optional notes. Notes persist in browser local storage.
8. Use Compare for measurements, profiles, slope histograms and candidate rankings. All successfully loaded Earth windows remain selectable, including windows marked Coverage review required. Their terrain score and ranking are withheld; measurements describe the full DEM window and may include water or bathymetry. New searches record all 25 coarse land-screen sample locations for inspection. Earth candidates can also be explored in 3D. What separates these candidates? explains the gap between the first two displayed scores, the selected candidate’s rank/gap, equal-score shared ranks, and any relief-distance cap. Only eligible loaded Earth windows are ranked; this is not an exhaustive Earth search. Point gaps are not confidence levels.
9. Export the Markdown comparison report, evidence JSON or field notebook. Each contains source information; reports include the method, candidate origin, recorded survey coverage and interpretation limits.
10. Use Sources to inspect exact provenance, citations and assumptions.
11. Open Planet data for original global maps, coordinate frames, physical dimensions and dated environmental records. Same physical scale retains one metres-to-scene conversion for all three bodies; Fit body to view changes the camera only. The global Elevation palette uses source-datum heights at 1° sampling and no exaggerated relief. Expand map gives the globe the full workspace width; Restore layout brings the mission panels back. Controls and evidence sit outside the 3D stage. Earth shows original NASA image brightness by default; Brighten Earth is an optional, labelled display adjustment that never affects elevation, temperature, water-map colours or the original image files.
12. Under Earth in Planet data, show dated NASA GIBS satellite imagery, daytime land-surface temperature or the historical annual water mask on the globe. The default dates are bundled original NASA renditions; other verified dates fetch NASA directly. Use the original legend and respect missing data. Temperature colours are not decoded into quantitative measurements.
13. Inspect NASA POWER 2025 skin/air temperatures, 10 m wind, pressure and precipitation for the selected Earth candidate. EONET initially shows an explicitly dated bundled storm catalogue; Refresh from NASA requests current catalogue metadata. Catalogue points are approximate dated reports, not current storm boundaries or evidence that other areas have no storms.

Scientific globe and environmental evidence

The active imagery comes unchanged from NASA Blue Marble Next Generation (November 2004), NASA/JPL Viking Mars imagery, and the USGS/NASA LROC lunar mosaic. The published delivery resolutions are 5400×2700, 1440×720 and 1024×512. The Mars texture’s exact MDIM revision/control accuracy is not stated by NASA; we verify global layout without claiming survey-grade texture registration.

Numerical geometry uses WGS84 plus NOAA ETOPO1 topography/bathymetry, NASA MOLA absolute radius, and NASA LOLA radial height. The global mesh is a 1° sample, not the full-resolution source. Earth MSL heights are applied to WGS84 without a geoid correction: a disclosed approximation. Ocean terrain is bathymetry, not a moving water surface. Existing finer study-window analysis remains separate. The renderer does not claim a current inertial orientation or sunlight model.

Earth environmental records are historical model/reanalysis, not live weather. Mars pressure comes from actual MEDA sol 29 PDS data with original CSV/XML and hashes retained; the three channels are separate and numerical uncertainties are unavailable. Current global Mars storms, local lunar ice detection and local water classification are not invented. Missing data remain missing.

Reproduction and source detail: imagery, global geometry, environment. Re-fetch GIBS references with node scripts/fetch-earth-layer-reference.mjs.

Exploration tasks

The Exploration task selector changes the comparison, not just the wording:

Task	Slope weight	Relief weight	Measures shown in Compare
Terrain exploration (default)	0.65	0.35	Relative relief, mean slope, 90th percentile slope
Navigation training	0.85	0.15	Ground under 10°, ground at 20° or steeper, 90th percentile slope
Field operations	0.50	0.50	Ground under 5°, relative relief, mean slope
Thermal environment study	0.65	0.35	Relative relief, mean slope; temperature stays separate evidence

Selecting a task re-ranks the loaded candidates, shows the reason for the preset under the selector, and adds a task panel to Compare with task-relevant measures, how the analog can help and what it cannot tell you. An unchecked characteristic stays off. Exports record the task and its preset. The presets are disclosed, project-defined weights for the same terrain-only index; they are not calibrated against field or mission results.

Actual datasets

• NASA MGS MOLA MEGDR: Jezero, Gale / Bradbury Landing and Eberswalde delta area, Mars.
• NASA LRO LOLA LDEM_75S_240M: Shackleton, Cabeus center and Nobile center, Moon.
• Skadi mixed-source elevation tiles: Wadi Rum, Al Wahbah, Amboy and Lanzarote.
• All study windows: 32 × 32 km; 65 × 65 samples; 500 m sampling; 1 km slope baseline.
• Additional Mars windows retain native MOLA observation-count summaries. Zero-count source bins contain archived interpolated topography; native posting is not measurement accuracy. These are broad terrain windows, not rover hazard maps.
• Cabeus/Nobile are crater-center windows, not the LCROSS impact or historical VIPER western-rim site. New targets have no bundled local temperature product; Jezero MEDA and Shackleton Diviner data are not transferred to them.
• Lanzarote is deliberately excluded from the inland ranking because its window includes coastal bathymetry. Amboy is an established reference site, not a new discovery.

See DATA.md for exact products, projections, byte ranges, source hashes, reproduction commands and formulas. See ASSETS.md for visual credits, licenses and the original procedural hardware model APIs.

See SEARCH-DATA.md for runtime terrain retrieval, PNG decoding, manual/automatic search contracts, real network checks and coverage limits. LANDMASK.md records both coastline assets and their hashes.

On-demand windows use RGB-encoded Mapzen / Tilezen Terrarium mixed-source DEM tiles. Direct PNG decoding avoids color conversion, retains negative land elevations, and samples pixel centers across tile seams. Search supports centers between 84° S and 84° N. A 5 × 5 Natural Earth land-mask screen excludes windows containing mapped non-land samples from scores and inland ranking, while retaining access to Compare, provenance, exports and 3D inspection. It is a coarse screen, not proof that every point is dry land, and it does not reliably classify inland lakes.

Scope and scientific interpretation

The resemblance index compares coarse slope distributions and relative relief. It is a transparent terrain heuristic, not a probability of mission suitability. It does not establish mineral, thermal, ice, biological or engineering equivalence. Material descriptions are qualitative context with linked sources.

Terrain elevation geometry comes from the sourced DEM. Fine ground textures, decorative rocks, sky colors and lighting are visual treatments. The explorer avatar is enlarged for navigation; its speed, gait and the lander arrival sequence are not validated robotics or flight simulations. Decorative rocks do not identify actual planetary hazards. Lunar lighting does not reconstruct permanent shadow. Coarse terrain resemblance supports a shortlist for further investigation; it is not a landing-site, permanent-base or field-safety certification.

Verification

npm ci
npm run check
npm test

Tests cover analytical slope calculations, cross-datum relief comparison, interpolation, polar geolocation, data eligibility, model geometry, full expedition state/movement logic, DOM workflows and scientific exports. DOM and expedition logic tests use a DOM environment and a renderer test double; they are not visual or GPU-rendering tests. A desktop-browser visual review is still needed when browser automation is available.

Structure

• dist/index.html, dist/styles.css: responsive application shell and design.
• dist/js/app.js: mission workflow, comparisons, notebook, dialogs and exports.
• dist/js/globe.js: interactive planet globe and hardware preview.
• dist/js/expedition.js: real-time scene, deployment, controls, minimap and layers.
• dist/js/models.js: original six-legged robot, cargo rocket and rock meshes.
• dist/js/science.js: source catalogue, sampling, geolocation and comparison.
• dist/js/terrain-search.js: on-demand terrain, PNG decoding, place lookup and bounded grid surveys with cancellation and source hashes.
• dist/js/search-ui.js, dist/search.css: manual and automatic discovery workflow.
• dist/js/search-cache.js: validation and local caching of recent search windows.
• dist/data/: ten sourced terrain subsets and source summary.
• scripts/fetch-terrain.mjs: reproducible data retrieval and subset pipeline.
• .openai/hosting.json: private Sites project and static hosting configuration.

Use of public space-agency data does not imply agency endorsement.

Surface thermal context

Choose Thermal environment study, then open Compare for source-derived Earth, Moon and Mars temperature records. The 3D Temperature layer distinguishes spatial lunar data from coarse Earth daily-mean context. Rover point observations are not extended over an entire crater. Thermal evidence retains its own statistic and time coverage and does not change the terrain-only ranking. See THERMAL.md, THERMAL-EARTH.md and THERMAL-PLANETARY.md for definitions and limits.